What does this skill do?

Skill's Continuous Session Learning captures the tacit knowledge generated during your Claude Code sessions and turns it into reusable skills. Using an auto-close hook, it extracts patterns such as error resolution, user corrections, workarounds, and debugging techniques, saving them to disk so that each session improves the performance of subsequent ones.

Self-Locking Hook
Configure the "Stop" hook in settings.json to extract patterns at the end of each session without adding latency.
Heal Learned Skills
Review and refine the patterns extracted in ~/.claude/skills/learned/, by removing noise or adjusting `ignore_patterns`.
Adjust Extraction Thresholds
Configure the minimum session length, the extraction threshold, and the categories of detectable patterns.
Troubleshoot Failed Extractions
Investigate why certain sessions aren't generating skills, and adjust the settings.

Usage examples

🪝 Enable automatic extraction
Configure the Stop hook to automatically extract patterns at the end of each Claude Code session.
🔍 Review skills learned
Show me which patterns have been extracted in ~/.claude/skills/learned/ and help me filter out the irrelevant ones.
⚡ Immediate manual extraction
Run /learn to force pattern extraction on the current conversation without waiting for it to end.
⚙️ Adjust sensitivity
Set the extraction threshold to "high" and add "external_api_issues" to the "ignore_patterns" to reduce noise.

Features

Hook Stop with no latency The evaluator runs only once at the end of the session, without affecting performance during the active conversation.
5 categories of patterns Identifies error resolution, user fixes, workarounds, debugging techniques, and project conventions.
Flexible Configuration Define the minimum session length, extraction threshold, auto_approve, and custom ignore_patterns lists.
Manual extraction using /learn Forces the immediate extraction of patterns from the current conversation without waiting for the session to end.
Structured Departure by Skipper Generate individual files containing the type, description, context, and recommendation for each learned pattern.

Frequently asked questions

No. The skill uses a shell hook (Stop) that runs locally when the Claude Code session closes. It does not require any external MCP.
No. The Stop hook runs only once at the end of the session, when the transcript is already complete. It does not introduce any latency during the active conversation.
If the session does not reach the configured minimum length or does not contain relevant patterns, the hook terminates without generating any files.
Yes. With `auto_approve: false` in `config.json`, candidate patterns remain in a pending review status until you manually approve them.
Continuous Session Learning — Pattern Extraction with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.